Well no that does not sound like denormalization. That would only happen if changing something lead to a very low (close to nothing) signal appearing at the input to some types of plugin. And I'm not even sure it's much of a possibility nowadays. Sorry if I sent you on a wild goose chase I had a different picture in my head for what you were doing.
There have been occasional reports of CPU errors and hardware inserts over the years. Some triggered by adding the insert, especially at very low buffer levels, and some reports that sound like yours when removing them. I could not find the one I was thinking of but found this.
https://duc.avid.com/showthread.php?t=380877 and
https://duc.avid.com/showthread.php?t=401017
I'd hope other folks might have ideas, the rarity of this makes me wonder if corrupt sessions could be happening to some people. And since I am pretty sure this has happened to some folks with UAD plugins in the past (can't quickly find those posts now) that may be a factor. Are you using UAD plugins on that master bus? In that session? Is your IO buffer set as large as it can be? What happens if you remove those plugins? Or try a quick brute force test of move all UAD plugins out of the plugin folder. What happens if you move the insert to the other extreme end of the chain? Just for curiosity I might try disabling latency compensation (a wild guess based on the high latency of UAD and other plugins folks are likely to use on a master bus... but problem triggers *might* be very routing dependent as well, especially if that routing is forcing latency comp to do much.).